US10229504B2 - Method and apparatus for motion estimation - Google Patents

Method and apparatus for motion estimation Download PDF

Info

Publication number
US10229504B2
US10229504B2 US14/777,384 US201414777384A US10229504B2 US 10229504 B2 US10229504 B2 US 10229504B2 US 201414777384 A US201414777384 A US 201414777384A US 10229504 B2 US10229504 B2 US 10229504B2
Authority
US
United States
Prior art keywords
image
interest
integral
region
integral image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active, expires
Application number
US14/777,384
Other languages
English (en)
Other versions
US20160035104A1 (en
Inventor
Petronel Bigioi
Peter Corcoran
Piotr Stec
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tobii Technology Ltd
Adeia Media Holdings LLC
Original Assignee
Fotonation Ireland Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fotonation Ireland Ltd filed Critical Fotonation Ireland Ltd
Priority to US14/777,384 priority Critical patent/US10229504B2/en
Assigned to FOTONATION LIMITED reassignment FOTONATION LIMITED ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: BIGIOI, PETRONEL, CORCORAN, PETER, STEC, PIOTR
Publication of US20160035104A1 publication Critical patent/US20160035104A1/en
Application granted granted Critical
Publication of US10229504B2 publication Critical patent/US10229504B2/en
Assigned to TOBII TECHNOLOGY LIMITED reassignment TOBII TECHNOLOGY LIMITED CHANGE OF NAME (SEE DOCUMENT FOR DETAILS). Assignors: FOTONATION LIMITED
Assigned to TOBII TECHNOLOGIES LIMITED reassignment TOBII TECHNOLOGIES LIMITED CHANGE OF NAME (SEE DOCUMENT FOR DETAILS). Assignors: FOTONATION LIMITED
Assigned to BANK OF AMERICA, N.A., AS COLLATERAL AGENT reassignment BANK OF AMERICA, N.A., AS COLLATERAL AGENT SECURITY INTEREST Assignors: ADEIA GUIDES INC., ADEIA HOLDINGS INC., ADEIA IMAGING LLC, ADEIA INC. (F/K/A XPERI HOLDING CORPORATION), ADEIA MEDIA HOLDINGS INC., ADEIA MEDIA LLC, ADEIA MEDIA SOLUTIONS INC., ADEIA PUBLISHING INC., ADEIA SEMICONDUCTOR ADVANCED TECHNOLOGIES INC., ADEIA SEMICONDUCTOR BONDING TECHNOLOGIES INC., ADEIA SEMICONDUCTOR INTELLECTUAL PROPERTY LLC, ADEIA SEMICONDUCTOR SOLUTIONS LLC, ADEIA SEMICONDUCTOR TECHNOLOGIES LLC, ADEIA SOLUTIONS LLC, ADEIA TECHNOLOGIES INC.
Assigned to ADEIA MEDIA HOLDINGS LLC reassignment ADEIA MEDIA HOLDINGS LLC ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: TOBII TECHNOLOGIES LTD
Assigned to ADEIA MEDIA HOLDINGS INC. reassignment ADEIA MEDIA HOLDINGS INC. CONVERSION Assignors: ADEIA MEDIA HOLDINGS LLC
Active legal-status Critical Current
Adjusted expiration legal-status Critical

Links

Images

Classifications

    • G06T7/2006—
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/223—Analysis of motion using block-matching
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
    • H04N19/51—Motion estimation or motion compensation
    • H04N19/53—Multi-resolution motion estimation; Hierarchical motion estimation
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
    • H04N23/60—Control of cameras or camera modules
    • H04N23/68—Control of cameras or camera modules for stable pick-up of the scene, e.g. compensating for camera body vibrations
    • H04N23/681—Motion detection
    • H04N23/6811—Motion detection based on the image signal
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
    • H04N23/60—Control of cameras or camera modules
    • H04N23/68—Control of cameras or camera modules for stable pick-up of the scene, e.g. compensating for camera body vibrations
    • H04N23/681—Motion detection
    • H04N23/6812—Motion detection based on additional sensors, e.g. acceleration sensors
    • H04N5/23229—
    • H04N5/23254—
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20004—Adaptive image processing
    • G06T2207/20012—Locally adaptive
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20068—Projection on vertical or horizontal image axis

Definitions

  • the present invention relates to a method and apparatus for motion estimation.
  • WO2008/151802 (Reference: FN-174) and WO2011/069698 (Reference: FN-352) disclose correlating profiles for respective image frames in a video sequence to determine relative movement between the image frames—the movement comprising either camera movement or subject movement. Providing a global measure of frame-to-frame motion however, has limited application.
  • U.S. Pat. No. 8,200,020 B1 discloses a computing device selecting a source tile from a source image. From the source tile, the computing device may select a first rectangular feature and a second rectangular feature. Based on the first and second rectangular features, the computing device may calculate a source feature vector. The computing device may also select a search area of a target image, and a target tile within the within the search area. Based on the target tile, the computing device may calculate a target feature vector. The computing device may determine that a difference between the source feature vector and the target feature vector is below an error threshold, and based on this determination, further determine a mapping between the source image and the target image. The computing device may then apply the mapping to the source image to produce a transformed source image.
  • U.S. Pat. No. 6,809,758 discloses stabilizing a motion image formed using a sequence of successive frames which includes calculating a motion vector field between adjacent frames; forming a motion vector histogram from horizontal and vertical components of the motion vector field; applying a threshold to the motion vector histogram to produce a thresholded motion vector histogram; generating average horizontal and vertical motion components from the thresholded motion vector histogram; filtering the average horizontal and vertical motion components over a number of frames to identify unwanted horizontal and vertical motion components for each of the frames; and stabilizing the image sequence by shifting each frame according to the corresponding unwanted horizontal and vertical motion.
  • a method of estimating motion between a pair of image frames of a given scene according to claim 1 .
  • This aspect of the invention employs an integral image derived from each image frame to determine relative motion between image frames at a number of levels of a hierarchy of image regions.
  • the motion between corresponding regions is not found directly using image correlation but with integral image profiles.
  • An integral image profile is a linear array containing sums of intensities of all pixels within columns or rows from a region of interest of an image. Integral image profiles from corresponding regions are correlated in order to find displacement between regions.
  • each of the levels of the hierarchy is divided into one or more regions so that the number of regions increases for each level down the hierarchy, e.g. at a base level, the image is divided into 16 ⁇ 16 regions, the next level up, has 8 ⁇ 8, next 4 ⁇ 4 and so on.
  • sampling of the integral image information is scaled, so that each level is sampled at twice the resolution of the level above, so providing an ever finer estimate of motion for successively more localised regions of an image.
  • Embodiments of the invention optimize the building of the integral profiles for each block of the pyramid and so provide an efficient way of performing hierarchical motion estimation that minimizes the amount of memory and memory bandwidth requirements as well as reducing computational complexity.
  • a method of estimating motion between a pair of image frames of a given scene According to a second aspect there is provided a method of estimating motion between a pair of image frames of a given scene.
  • a third aspect there is provided a method of estimating motion between a pair of image frames of a given scene according to claim 17 .
  • motion estimation is started one or more levels below a root level of the hierarchy.
  • FIG. 1 is a block diagram of an image processing apparatus arranged to perform motion estimation according to an embodiment of the present invention
  • FIG. 2 is a flow diagram illustrating generally a method of motion estimation according to an embodiment of the present invention
  • FIG. 3 shows the integral image pixels used to generate an integral image profile for a first iteration of the method of FIG. 2 ;
  • FIG. 4 illustrates a target image (T) displaced relative to a reference image (R);
  • FIG. 5 illustrates exemplary integral image profiles for a pair of displaced images such as shown in FIG. 4 ;
  • FIG. 6 shows displaced regions of interest (ROIs) at a second iteration of the method of FIG. 2 ;
  • FIG. 7 shows the pixels required to build integral image profiles for the top and bottom left ROIs of a reference image (R) at a second iteration of the method of FIG. 2 ;
  • FIG. 8 shows displaced regions of interest (ROIs) at a third iteration of the method of FIG. 2 ;
  • FIG. 9 illustrates an interpolated integral image profile
  • FIGS. 10 and 11 illustrate the calculation of sub-pixel displacement between profiles based on mean absolute error (MAE);
  • FIG. 12 shows an object covering significant part of an image frame
  • FIG. 13 shows a profile indicating MAE with local minima for the image of FIG. 12 ;
  • FIG. 14 shows a displacement matrix of motion vectors produced according to the method of FIG. 2 ;
  • FIG. 15 illustrates a method for selecting vectors from the matrix for use in calculating a global transformation matrix
  • FIG. 16 shows a selection mask for the matrix of FIG. 14 produced by the method of FIG. 15 ;
  • FIGS. 17-19 illustrate a non-uniform sub-division of a region of interest from one level of the hierarchy to the next.
  • FIG. 1 there is shown schematically an image processing device 10 for performing motion estimation according to an embodiment of the present invention.
  • the device includes a bus 12 which allows functional processing modules 14 - 22 (shown below the bus) to read and write information to memory 24 - 28 (shown above the bus). It should be noted that the modules 14 - 22 can incorporate local memory to facilitate internal processing.
  • Image frames are acquired via a down sampler (DS) 14 from an image sensor (not shown).
  • the down sampler 14 may for example be a Gaussian down-sampler of the type provided by Fujitsu.
  • the down-sampled image is fed to an integral image (II) generator (GEN) 16 which writes the II to memory 24 .
  • II integral image
  • Calculation of integral image is well known and was originally disclosed by Viola, P. and Jones, M. in “Rapid Object Detection using a Boosted Cascade of Simple Features”, Computer Vision and Pattern Recognition, 2001, Volume 1.
  • Integral images are typically used in identifying objects such as faces in images, such as disclosed in WO2008/018887 (Reference: FN-143).
  • an intensity version of the original image is required to provide an integral image. This could be a grey scale version of the image, or it could be any single plane of a multi-plane image format, for example, RGB, LAB, YCC etc.
  • a hierarchical registration engine (HRE) 18 reads integral image information for a pair of frames from memory 24 and generates a displacement map 26 for the image pair as will be described in more detail below.
  • a CPU module 20 running an application program can then use displacement maps 26 for successive image frames to provide configuration information 28 required, for example, by a graphics distortion engine (GDE) 22 of the type described in WO 2014/005783 (Reference: FN-384) to provide image stabilization within a video sequence.
  • GDE graphics distortion engine
  • the HRE module 18 does not use the video frame directly but rather uses integral image information calculated from a down-sampled representation of the image frame.
  • the HRE module 18 requires buffering of integral image information for two frames in memory 24 , using one set of image information for a reference frame and calculating the displacement of region(s) of interest (ROI) within a target frame relative to the reference frame.
  • ROI region(s) of interest
  • the reference frame can alternate temporally with the target frame, so that it precedes the target frame and then succeeds the target frame.
  • the HRE module 18 performs a hierarchical search in order to find motion vectors for the regions of interest at each level of the hierarchy. It is coarse-to-fine approach where the search is performed first on integral image information for the complete image frame at a largest sub-sampling interval. Then the frame is split into a plurality of regions and the motion estimate for the complete frame is used as an initial guess for local motion in the individual regions; and so on down through the hierarchy.
  • the module 18 builds an integral image profile for each of the reference frame (R) and the target frame (T) based on integral image data 24 retrieved in memory, step 32 .
  • An integral image profile is an array that contains in each element, a sum of all pixel intensities in a corresponding swath, column or row—depending on the search direction, of a region of interest of an image.
  • the integral image profile is stored locally within the HRE module 18 , although it could be written back into general purpose memory if required.
  • calculating the integral image profile for a given region of interest of the hierarchy involves sub-sampling the integral image along the first row of the ROI and subtracting these values R 1-x from their sub-sampled values R 2-x along the last row of the ROI, the top row values marking the top-left corner and the bottom row values marking the bottom-right corner of each swath providing a value within an integral image profile.
  • R 1-1 0
  • the profile value for the first column is simply R 2-1
  • the profile value for the next column is simply R 2-2 -R 2-1 and so on across the width of the image.
  • only the bottom row of integral image information needs to be sampled at the top level of the hierarchy to generate the integral image profile at this level. (A similar approach applies for determining vertical displacement.)
  • FIG. 4 shows a pair of images T, R which are horizontally and vertically displaced and FIG. 5 shows horizontal integral image profiles T, R for these images.
  • These integral image profiles can be readily correlated to determine the displacement of the target frame from the reference frame. (Again, the same operation is performed in order to find the vertical displacement but in this case, profiles are built by summing rows of the images.)
  • each of the levels of hierarchy is sampled at 1 ⁇ 2 the resolution of the level lying directly below, with the coarsest full image version at the top of the hierarchy and the finest at the bottom.
  • motion in integral image samples found at one level of the hierarchy, is multiplied by 2 and set as an initial estimate to the level below, its nominal inaccuracy in the absence of local movement being ⁇ 1 sample in each direction.
  • a low pass filter can be applied to the matrix to reduce the influence of outliers.
  • a row of displacement values A-D from one level are upscaled to produce a row of start displacement values a-h for the next level:
  • each ROI is split into 4 new ROIs when going to the next level of the hierarchy, step 34 , FIG. 2 .
  • the subdivision level can vary and could be even dynamic, with variable or non-uniform ROI sizes.
  • FIG. 6 shows a layout of the ROIs after a first iteration of the method—reference frame (R) is shown on the right and the target frame (T) on the left.
  • reference frame (R) is shown on the right and the target frame (T) on the left.
  • all four ROIs are initialized with the same motion, step 36 , because it comes from a single vector determined from the analysis of the integral image information for the top level of the hierarchy.
  • the image information for regions of interest of the target image (T) is taken from samples shifted relative to the samples of the reference image (R) according to the motion determined at the previous higher level of the hierarchy.
  • profiles and displacements of the target image relative to the reference image are determined for each of the 4 regions of interest shown in FIG. 6 , step 32 .
  • the integral image profiles are built by sampling integral image points along the top and bottom rows of each of the 4 regions of the image.
  • the integral image information for the displaced regions of the target image is sampled from locations shifted according to the displacement detected for the region of interest bounding the current regions from the higher level of the hierarchy.
  • motion for the second level can be determined again at step 32 , before the process is repeated for the next level of the hierarchy at steps 38 , 34 and 36 .
  • integral image information is sub-sampled and so downscaled 4 times compared to the original integral image resolution.
  • To calculate the horizontal integral image profile every 4 th pixel from the bottom most line of the integral image is sampled. By calculating differences between successive samples, integral image profile values are obtained. For an original frame size of 640 ⁇ 480, the top level of the pyramid would require 160 values for each image.
  • the blocks from the next lower level of the pyramid require every second line of pixels from the integral image to be sampled in order to calculate the required profiles. For example, to calculate profile from the bottom left quarter of the integral image at 50% of the original integral image resolution, every second pixel from the two lines L 0 and L 1 , are read from the integral image as shown in FIG. 7 . Thus, for a 640 ⁇ 480 image, up to 320 pixels of integral image information per ROI are required, which is still 60 times less bandwidth than traditional methods require.
  • This method of building the profiles allows for arbitrary location of image blocks within a target image and directly supports any integer downscaling factor of the original image without needing any additional processing.
  • step 40 it is also possible to determine sub-pixel displacement, step 40 .
  • Aligning profiles with sub-pixel precision allows, for example, low resolution image information, for example, VGA to determine precise motion within a high resolution image, for example, Full HD.
  • FIG. 9 shows original samples (circles) from a first profile interpolated to provide a continuous line profile. Samples from a second profile shifted by 0.5 of a pixel are shown as crossed. The displacement (error) is calculated as a mean absolute error (MAE) between values represented by the crosses and values of the blue line taken at corresponding locations. To save computation time, not all the values from the blue line are calculated. The function is evaluated only at the positions where error between two profiles must be evaluated.
  • MAE mean absolute error
  • FIG. 10 shows how the error changes with the displacement between profiles.
  • the point marked with 0 is the initial position for profiles that are aligned up to 1 pixel.
  • the search starts with the initial step which is ⁇ 0.5 of the search range. Error values evaluated at those positions are compared and the current position is moved to the location of the lesser error.
  • the operation is repeated for a fixed number of iterations that define required accuracy. To achieve 1/256 of a pixel accuracy, 8 iterations are used. The last steps of the search are shown in FIG. 11 and this determines that the displacement between the profiles is between 0.7 and 0.71 pixels.
  • FIG. 13 illustrates an MAE function for the integral images profiles for the top level of the hierarchy for the image of FIG. 12 .
  • the function is calculated by shifting one of the profiles in respect to the other and calculating a mean absolute error between displaced elements of the profiles.
  • the function contains a local minimum as well as the global minimum. The global minimum reflects the most dominant motion, but the local minimum, in this case, reflects the object motion.
  • Calculating the absolute difference of the profiles shifted by the location of the minima indicates the location of the moving object. This shows which part of the profile belongs to which object from the scene. This allows multiple motion values to be returned from the single profile correlation such as in step 32 of FIG. 2 , and, as a consequence, allows for more accurate initialization of the underlying blocks from the lower level of the hierarchy.
  • the motion in the right part of the profile from the top level of the pyramid reflects different motion than the left part.
  • the ROIs 2 and 4 can be now initialized with more appropriate displacement values.
  • FIG. 17 which illustrates a region of interest (ROI) in which an object ( 0 ) towards the centre of the region is moving relative to the background (B).
  • FIG. 18 shows two integral profiles, typical of those determined for horizontal displacement in such a region of interest.
  • FIG. 19 is a two dimensional map illustrating the error value along the profiles versus horizontal displacement.
  • the horizontal dimension is the length of the profiles while vertical dimension is the displacement.
  • the black top-left and bottom-right corners are due to non-overlapping parts of the profiles and need not taken into consideration. It can be observed that there is line of minimum error (L) formed for certain displacements and this line can be found with some constrained optimization methods, such as linear or quadratic programming with geometric constraints or active contour methods like snakes or level-sets.
  • the vertical position of the line L indicates displacement. Where the line is near horizontal it denotes a moving object or background and sloped sections of the line denote uncertainty areas.
  • the two sloped sections are used to find subdivision points H 1 , H 2 , in FIG. 17 for the region of interest. Again, the motion in the vertical direction and sub-divisions are found analogously to provide subdivision point V 1 in FIG. 17 .
  • the region of interest is divided for the lower level of the hierarchy into 3 horizontal regions and 2 vertical regions.
  • a built in motion sensor can provide a good way to reduce the number of levels required in the image hierarchy employed in the embodiment of FIG. 2 by providing an initial guess for the search and so avoid needing to calculate motion on all the levels of the hierarchy;
  • the top level displacement calculation can be omitted from the embodiment illustrated in FIG. 2 , and the motion calculated from the sensor measurements used as an initial guess for the second level (and possibly subsequent levels) for example as indicated in FIG. 6 .
  • the number of hierarchy levels that are needed to supplement the motion sensor(s) depends on the image size and the sensor accuracy. For example, if a sensor can provide accuracy + ⁇ 3 pixels, at least two levels of hierarchy with a search radius of + ⁇ 2 pixels at each level are required.
  • a displacement matrix comprising local motion vectors (each indicating local horizontal and vertical displacement, potentially with sub-pixel accuracy) such as illustrated in FIG. 14 is provided by the HRE module 18 and can be written to memory 26 .
  • a least squares or equivalent approach could be used. This approach would be sufficient if the displacement matrix contained relatively small measurement errors with a Gaussian distribution. Unfortunately, this is almost never the case.
  • the motion vectors can be invalid in many ways, for example, they can contain local motion that comes from a moving object, rather than camera motion, or they could be erroneous due to lack of detail in a scene being imaged or repeating patterns that interfere with the correlation process. As such, using raw displacement matrix information determined for ROIs within a scene directly to provide a geometrical transformation would be likely to produce poor results.
  • Embodiments of the present invention reduce the computational complexity of matrix estimation by several orders of magnitude, with predictable execution time and providing repeatable results as explained below.
  • the values of the motion vectors from FIG. 14 are first quantized into integer values, step 150 .
  • the level of quantization depends on the desired accuracy and the expected range of motion within the frame. In the simplest case, the quantization might be performed as rounding to the nearest integer value, but other quantization bins can be used.
  • a comparagram is built. This is 2D histogram in which each dimension represents the quantized motion in horizontal and vertical direction respectively and the value of the comparagram bin shows frequency of vectors sharing the same quantized motion values in both horizontal and vertical directions.
  • step 154 a maximum bin value within the comparagram is found.
  • the position of the maximum becomes a seed for a growing a region connecting neighbouring bins based on the similarity of their value to the value of the seed, step 156 .
  • All motion vectors within the displacement matrix that fall into marked bins are selected for motion estimation, step 158 .
  • the final motion estimation can be performed using standard least squares method, step 160 .
  • a sample selected vectors mask for the matrix of FIG. 14 is shown in FIG. 15 . This compares favourably with results provided by for example RANSAC albeit employing more rationalised and deterministic processing resources.
  • a Reduced Integral Image is stored in memory for every second acquired image.
  • Each such image is used as the reference image knowing that the required integral image profiles can be built from the samples coinciding with the boundaries illustrated in FIG. 8 RHS.
  • a complete II is stored for the alternate target images as the boundaries used for regions of interest vary according to the displacement calculated for higher levels of the hierarchy as can be seen for FIG. 8 LHS.
  • the complete integral image can of course be used by other processing modules including a face detector (not shown) and as disclosed in WO2008/018887 (Reference: FN-143), such detectors do not always require an integral image for every frame—thus embodiments of the present invention employing an RII do not necessarily impose a greater processing burden on a device which might already be performing face detection.
  • the generator 16 alternately writes to memory 24 , a full Integral Image (frame N) and a Reduced II (frame N+1); then II (frame N+2) and RII (frame N+3).
  • the HRE module 18 uses II(N) and RII(N+1) from memory 24 to produce the displacement map for frame N+1; and then uses RII(N+1) and II(N+2) from memory 24 to produce the displacement map for frame N+2.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
US14/777,384 2013-03-18 2014-03-14 Method and apparatus for motion estimation Active 2034-10-28 US10229504B2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US14/777,384 US10229504B2 (en) 2013-03-18 2014-03-14 Method and apparatus for motion estimation

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
US201361803115P 2013-03-18 2013-03-18
PCT/EP2014/055125 WO2014146983A2 (en) 2013-03-18 2014-03-14 A method and apparatus for motion estimation
US14/777,384 US10229504B2 (en) 2013-03-18 2014-03-14 Method and apparatus for motion estimation

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2014/055125 A-371-Of-International WO2014146983A2 (en) 2013-03-18 2014-03-14 A method and apparatus for motion estimation

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/298,657 Continuation US10587806B2 (en) 2013-03-18 2019-03-11 Method and apparatus for motion estimation

Publications (2)

Publication Number Publication Date
US20160035104A1 US20160035104A1 (en) 2016-02-04
US10229504B2 true US10229504B2 (en) 2019-03-12

Family

ID=50289660

Family Applications (2)

Application Number Title Priority Date Filing Date
US14/777,384 Active 2034-10-28 US10229504B2 (en) 2013-03-18 2014-03-14 Method and apparatus for motion estimation
US16/298,657 Active US10587806B2 (en) 2013-03-18 2019-03-11 Method and apparatus for motion estimation

Family Applications After (1)

Application Number Title Priority Date Filing Date
US16/298,657 Active US10587806B2 (en) 2013-03-18 2019-03-11 Method and apparatus for motion estimation

Country Status (6)

Country Link
US (2) US10229504B2 (de)
EP (2) EP2826022B1 (de)
JP (1) JP6022734B2 (de)
KR (1) KR101613176B1 (de)
CN (2) CN107248168A (de)
WO (1) WO2014146983A2 (de)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10499867B2 (en) * 2018-01-08 2019-12-10 Shenzhen Keya Medical Technology Corporation Method, storage medium, and system for analyzing image sequences of periodic physiological activities
US11786202B2 (en) 2018-01-08 2023-10-17 Shenzhen Keya Medical Technology Corporation Method, system, and medium for analyzing image sequence of periodic physiological activity

Families Citing this family (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8498695B2 (en) 2006-12-22 2013-07-30 Novadaq Technologies Inc. Imaging system with a single color image sensor for simultaneous fluorescence and color video endoscopy
EP3058510B1 (de) 2014-11-28 2017-12-20 FotoNation Limited Verfahren zur herstellung eines histogramms von orientierten gradienten
EP3230948B1 (de) 2015-04-30 2019-07-03 FotoNation Limited Verfahren und vorrichtung zur herstellung eines videostroms
EP3311364B1 (de) 2015-09-30 2018-12-26 FotoNation Limited Verfahren und system zur verfolgung eines objekts
KR102407624B1 (ko) * 2015-10-06 2022-06-10 삼성전자주식회사 전자 장치의 영상 처리 방법 및 그 전자 장치
EP4442222A3 (de) 2015-11-13 2024-12-18 Stryker Corporation Systeme und verfahren zur beleuchtung und bildgebung eines ziels
US11477382B2 (en) 2016-02-19 2022-10-18 Fotonation Limited Method of stabilizing a sequence of images
EP3417606B1 (de) 2016-02-19 2020-08-12 FotoNation Limited Verfahren zur stabilisierung einer bildersequenz
EP3469420A4 (de) * 2016-06-14 2020-02-12 Novadaq Technologies ULC Verfahren und systeme zur adaptiven bildgebung zur verstärkung eines schwachen lichtsignals bei einer medizinischen visualisierung
US10334152B2 (en) 2016-08-08 2019-06-25 Fotonation Limited Image acquisition device and method for determining a focus position based on sharpness
US10148943B2 (en) 2016-08-08 2018-12-04 Fotonation Limited Image acquisition device and method based on a sharpness measure and an image acquistion parameter
EP3580609B1 (de) 2017-02-10 2023-05-24 Stryker European Operations Limited Handhaltbare offenfeld-fluoreszenzbildgebungssysteme und verfahren
US10097757B1 (en) 2017-03-24 2018-10-09 Fotonation Limited Method for determining bias in an inertial measurement unit of an image acquisition device
CN107071421B (zh) * 2017-05-23 2019-11-22 北京理工大学 一种结合视频稳定的视频编码方法
FR3069688A1 (fr) * 2017-07-25 2019-02-01 Parrot Drones Dispositif electronique et procede d'estimation d'un deplacement d'une scene observee par un drone, appareil electronique de calcul d'une vitesse sol du drone, drone et programme d'ordinateur associes
DE102017216065A1 (de) * 2017-09-12 2019-03-14 Robert Bosch Gmbh Verfahren und Vorrichtung zum Bewerten von Bildern, Betriebsassistenzverfahren und Betriebsvorrichtung
CN107707913B (zh) * 2017-09-29 2019-12-17 福州大学 快速视频编码中防帧内误差传递方法
US10360662B2 (en) 2017-10-16 2019-07-23 Omnivision Technologies, Inc. Alignment of multiple camera images by matching projected one dimensional image profiles
US10726522B2 (en) * 2018-01-24 2020-07-28 Fotonation Limited Method and system for correcting a distorted input image
US11037316B2 (en) * 2018-03-30 2021-06-15 Canon Kabushiki Kaisha Parallax calculation apparatus, parallax calculation method, and control program of parallax calculation apparatus
JP7344660B2 (ja) * 2018-03-30 2023-09-14 キヤノン株式会社 視差算出装置、視差算出方法及び視差算出装置の制御プログラム
EP3547211B1 (de) * 2018-03-30 2021-11-17 Naver Corporation Verfahren zum trainieren eines cnn und klassifizierung einer aktion, die von einer person in einem eingegebenen video ausgeführt wird, mithilfe des besagten cnn
WO2021025375A1 (en) * 2019-08-06 2021-02-11 Samsung Electronics Co., Ltd. Apparatus and method for efficient regularized image alignment for multi-frame fusion
US11151731B2 (en) 2019-08-06 2021-10-19 Samsung Electronics Co., Ltd. Apparatus and method for efficient regularized image alignment for multi-frame fusion
US10983363B2 (en) 2019-09-19 2021-04-20 Fotonation Limited Method for stabilizing a camera frame of a video sequence
EP3985957B1 (de) * 2020-10-14 2022-11-30 Axis AB Verfahren und systeme zur bewegungssegmentierung
US11425412B1 (en) * 2020-11-10 2022-08-23 Amazon Technologies, Inc. Motion cues for video encoding
KR102524823B1 (ko) * 2020-12-04 2023-04-21 포항공과대학교 산학협력단 신경망 모델 기반 비디오의 움직임 특징 정보 추출 방법 및 분석장치
US11610338B2 (en) 2021-03-26 2023-03-21 Fotonation Limited Method of controlling a camera
CN113536570B (zh) * 2021-07-16 2022-09-27 中国石油大学(北京) 一种应力参数计算方法、装置、系统及计算机设备
CN115174811A (zh) * 2022-06-30 2022-10-11 Oppo广东移动通信有限公司 相机抖动检测方法、装置、设备、存储介质和程序产品

Citations (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6130912A (en) * 1998-06-09 2000-10-10 Sony Electronics, Inc. Hierarchical motion estimation process and system using block-matching and integral projection
US6459822B1 (en) 1998-08-26 2002-10-01 The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration Video image stabilization and registration
US6809758B1 (en) 1999-12-29 2004-10-26 Eastman Kodak Company Automated stabilization method for digital image sequences
US20050275727A1 (en) * 2004-06-15 2005-12-15 Shang-Hong Lai Video stabilization method
US20070237514A1 (en) * 2006-04-06 2007-10-11 Eastman Kodak Company Varying camera self-determination based on subject motion
WO2008018887A1 (en) 2006-08-11 2008-02-14 Fotonation Vision Limited Real-time face tracking in a digital image acquisition device
US20080159630A1 (en) * 2006-11-20 2008-07-03 Eitan Sharon Apparatus for and method of robust motion estimation using line averages
US20080273752A1 (en) * 2007-01-18 2008-11-06 Siemens Corporate Research, Inc. System and method for vehicle detection and tracking
WO2008151802A1 (en) 2007-06-14 2008-12-18 Fotonation Ireland Limited Fast motion estimation method
US20090058990A1 (en) * 2007-08-29 2009-03-05 Samsung Electronics Co., Ltd. Method for photographing panoramic picture
US20100111446A1 (en) * 2008-10-31 2010-05-06 Samsung Electronics Co., Ltd. Image processing apparatus and method
WO2011069698A1 (en) 2009-12-11 2011-06-16 Tessera Technologies Ireland Limited Panorama imaging
US8200020B1 (en) 2011-11-28 2012-06-12 Google Inc. Robust image alignment using block sums
US20130016180A1 (en) * 2011-07-13 2013-01-17 Hiroaki Ono Image processing apparatus, method, and program
US20130076921A1 (en) * 2011-09-26 2013-03-28 Christopher Owen Video Stabilization
WO2014005783A1 (en) 2012-07-03 2014-01-09 DigitalOptics Corporation Europe Limited A method and system for correcting a distorted input image

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20020018566A (ko) * 2000-09-01 2002-03-08 액티브 올가닉스 인코포레이티드 모노-아실-(리소)-글리세롤인지질로 국소 활성의 효능을강화하는 방법 및 이의 용도
US7616836B2 (en) * 2004-05-25 2009-11-10 Siemens Medical Solutions Usa, Inc. Method and system for motion compensation in a temporal sequence of images
US10080006B2 (en) * 2009-12-11 2018-09-18 Fotonation Limited Stereoscopic (3D) panorama creation on handheld device
CN101951464A (zh) * 2010-05-19 2011-01-19 上海稳像信息技术有限公司 基于积分图特征块匹配的实时视频稳像方法
WO2012066866A1 (ja) * 2010-11-17 2012-05-24 三菱電機株式会社 動きベクトル検出装置、動きベクトル検出方法、フレーム補間装置及びフレーム補間方法
CN102957869B (zh) * 2011-09-26 2016-08-17 斯凯普公司 视频稳定
CN102801982B (zh) * 2012-09-07 2015-03-18 浙江外国语学院 一种应用于视频压缩且基于块积分的快速运动估计方法

Patent Citations (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6130912A (en) * 1998-06-09 2000-10-10 Sony Electronics, Inc. Hierarchical motion estimation process and system using block-matching and integral projection
US6459822B1 (en) 1998-08-26 2002-10-01 The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration Video image stabilization and registration
US6809758B1 (en) 1999-12-29 2004-10-26 Eastman Kodak Company Automated stabilization method for digital image sequences
US20050275727A1 (en) * 2004-06-15 2005-12-15 Shang-Hong Lai Video stabilization method
US20070237514A1 (en) * 2006-04-06 2007-10-11 Eastman Kodak Company Varying camera self-determination based on subject motion
WO2008018887A1 (en) 2006-08-11 2008-02-14 Fotonation Vision Limited Real-time face tracking in a digital image acquisition device
US20080159630A1 (en) * 2006-11-20 2008-07-03 Eitan Sharon Apparatus for and method of robust motion estimation using line averages
US20080273752A1 (en) * 2007-01-18 2008-11-06 Siemens Corporate Research, Inc. System and method for vehicle detection and tracking
WO2008151802A1 (en) 2007-06-14 2008-12-18 Fotonation Ireland Limited Fast motion estimation method
US20090058990A1 (en) * 2007-08-29 2009-03-05 Samsung Electronics Co., Ltd. Method for photographing panoramic picture
US20100111446A1 (en) * 2008-10-31 2010-05-06 Samsung Electronics Co., Ltd. Image processing apparatus and method
WO2011069698A1 (en) 2009-12-11 2011-06-16 Tessera Technologies Ireland Limited Panorama imaging
US20130016180A1 (en) * 2011-07-13 2013-01-17 Hiroaki Ono Image processing apparatus, method, and program
US20130076921A1 (en) * 2011-09-26 2013-03-28 Christopher Owen Video Stabilization
US8200020B1 (en) 2011-11-28 2012-06-12 Google Inc. Robust image alignment using block sums
WO2014005783A1 (en) 2012-07-03 2014-01-09 DigitalOptics Corporation Europe Limited A method and system for correcting a distorted input image

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
European Patent Office, Extended European Search Report dated Jun. 9, 2016 for EP Application No. 15182504.9 /Pub. No. 2983136 titled "A Method and Apparatus for Motion Estimation",14 pages.
Nguyen, Viet Anh, and Yap-Peng Tan. "Fast block-based motion estimation using integral frames." IEEE signal processing letters 11, No. 9 (2004): 744-747. *
Sauer, Ken, and Brian Schwartz. "Efficient block motion estimation using integral projections." IEEE Transactions on circuits and systems for video technology 6, No. 5 (1996): 513-518. *
Viola, P. And Jones, M Rapid Object Detection using a Boosted Cascade of Simple Features,: Computer Vision and Pattern Recognition, 2001, vol. 1, pp. I-511 to I-518.
WIPO, International Preliminary Report on Patentability Issued Sep. 22, 2015 for PCT App. No. PCT/EP2014/055125 titled "A Method and Apparatus for Motion Estimation", 12 pages.
WIPO, International Search Report and Written Opinion dated Sep. 23, 2014 for PCT App. No. PCT/EP2014/055125 titled "A Method and Apparatus for Motion Estimation", 6 pages.

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10499867B2 (en) * 2018-01-08 2019-12-10 Shenzhen Keya Medical Technology Corporation Method, storage medium, and system for analyzing image sequences of periodic physiological activities
US10980502B2 (en) 2018-01-08 2021-04-20 Shenzhen Keya Medical Technology Corporation Method, storage medium, and system for analyzing image sequences of periodic physiological activities
US11786202B2 (en) 2018-01-08 2023-10-17 Shenzhen Keya Medical Technology Corporation Method, system, and medium for analyzing image sequence of periodic physiological activity

Also Published As

Publication number Publication date
US20160035104A1 (en) 2016-02-04
EP2983136A3 (de) 2016-07-13
EP2983136B1 (de) 2017-11-22
EP2826022B1 (de) 2015-10-21
KR101613176B1 (ko) 2016-04-29
US10587806B2 (en) 2020-03-10
EP2826022A2 (de) 2015-01-21
WO2014146983A2 (en) 2014-09-25
CN105051785A (zh) 2015-11-11
US20190273867A1 (en) 2019-09-05
JP2016514867A (ja) 2016-05-23
WO2014146983A3 (en) 2014-11-13
JP6022734B2 (ja) 2016-11-09
CN105051785B (zh) 2017-07-04
EP2983136A2 (de) 2016-02-10
KR20150119395A (ko) 2015-10-23
CN107248168A (zh) 2017-10-13

Similar Documents

Publication Publication Date Title
US10587806B2 (en) Method and apparatus for motion estimation
US11593654B2 (en) System for performing convolutional image transformation estimation
US9681150B2 (en) Optical flow determination using pyramidal block matching
US8958484B2 (en) Enhanced image and video super-resolution processing
US9990753B1 (en) Image stitching
US10311595B2 (en) Image processing device and its control method, imaging apparatus, and storage medium
US9536147B2 (en) Optical flow tracking method and apparatus
US8150197B2 (en) Apparatus and method of obtaining high-resolution image
JP4404095B2 (ja) テンプレートマッチングのための方法、その装置及びそのためのプログラムを記録した記録媒体
US9240048B2 (en) Depth estimation using three-dimensional epipolar data structures
US10621446B2 (en) Handling perspective magnification in optical flow processing
CN107507225B (zh) 运动目标检测方法、装置、介质及计算设备
US20130027775A1 (en) Image stabilization
US8903195B2 (en) Specification of an area where a relationship of pixels between images becomes inappropriate
US11954835B2 (en) Methods, devices, apparatuses, and media for image fusion utilizing images and LiDAR point clouds
US10692235B2 (en) Image processing apparatus and method for determining a depth of a pixel of a reference image
TW202338734A (zh) 用於處理影像資料的方法及影像處理器單元
CN115272519A (zh) 动态区域检测方法及存储介质
CN116156075A (zh) 视频中目标相对运动预测方法、装置和计算机设备
CN112784828A (zh) 基于方向梯度直方图的图像检测方法、装置、计算机设备
Sreegeethi et al. Online Video Stabilization using Mesh Flow with Minimum Latency
US8867843B2 (en) Method of image denoising and method of generating motion vector data structure thereof

Legal Events

Date Code Title Description
AS Assignment

Owner name: FOTONATION LIMITED, IRELAND

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:BIGIOI, PETRONEL;CORCORAN, PETER;STEC, PIOTR;SIGNING DATES FROM 20150827 TO 20150831;REEL/FRAME:036571/0359

STCF Information on status: patent grant

Free format text: PATENTED CASE

MAFP Maintenance fee payment

Free format text: PAYMENT OF MAINTENANCE FEE, 4TH YEAR, LARGE ENTITY (ORIGINAL EVENT CODE: M1551); ENTITY STATUS OF PATENT OWNER: LARGE ENTITY

Year of fee payment: 4

AS Assignment

Owner name: TOBII TECHNOLOGY LIMITED, IRELAND

Free format text: CHANGE OF NAME;ASSIGNOR:FOTONATION LIMITED;REEL/FRAME:070238/0774

Effective date: 20240820

AS Assignment

Owner name: TOBII TECHNOLOGIES LIMITED, IRELAND

Free format text: CHANGE OF NAME;ASSIGNOR:FOTONATION LIMITED;REEL/FRAME:070682/0207

Effective date: 20240820

AS Assignment

Owner name: BANK OF AMERICA, N.A., AS COLLATERAL AGENT, ILLINOIS

Free format text: SECURITY INTEREST;ASSIGNORS:ADEIA INC. (F/K/A XPERI HOLDING CORPORATION);ADEIA HOLDINGS INC.;ADEIA MEDIA HOLDINGS INC.;AND OTHERS;REEL/FRAME:071454/0343

Effective date: 20250527

AS Assignment

Owner name: ADEIA MEDIA HOLDINGS LLC, CALIFORNIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:TOBII TECHNOLOGIES LTD;REEL/FRAME:071572/0855

Effective date: 20250318

Owner name: ADEIA MEDIA HOLDINGS INC., CALIFORNIA

Free format text: CONVERSION;ASSIGNOR:ADEIA MEDIA HOLDINGS LLC;REEL/FRAME:071577/0875

Effective date: 20250408